Multimodal Spatiotemporal Power Flow Data Reserve Pool Network Prediction Method and Equipment

Abstract

This invention provides a method and device for predicting a multimodal spatiotemporal power flow data reservoir network, comprising: acquiring multimodal spatiotemporal power flow data, wherein the multimodal spatiotemporal power flow data includes time-series power flow data and four-dimensional coordinates of nodes; constructing a tree-like reservoir network model, using the tree-like reservoir network model to fuse features at different levels, and predicting the time-series power flow data to obtain future power flow data for each region; processing the four-dimensional coordinates of the nodes using four-dimensional fuzzy wavelet clustering to generate similar nodes; comparing the correlation of the predicted power flow data among the similar nodes, and calculating a data volatility index based on the correlation; and marking weak links in the microgrid group based on the data volatility index.

Description

Technical Field

[0001] This invention relates to the field of microgrid group weak link prediction technology, and in particular to a multimodal spatiotemporal power flow data storage pool network prediction method and equipment. Background Technology

[0002] Driven by the "dual carbon" goals, the construction of a new power system based on new energy sources has become inevitable. Microgrid clusters have become a research hotspot due to their ability to enable flexible and efficient application of distributed new energy sources and solve their grid connection problems. However, the high proportion of renewable energy integration and increased load volatility have made the operating conditions of microgrid clusters increasingly complex. Simultaneously, the deep application of information and communication technologies and distributed control technologies has transformed microgrid clusters into "information-physical-social" composite systems. While this improves energy efficiency, it has also led to an exponential increase in network attack paths. Attacks against microgrid clusters are showing a trend towards diversification of methods, greater precision in targeting, and more severe consequences. Without an effective situational awareness system and risk modeling and defense decision-making mechanisms, attacks targeting internal vulnerabilities cannot be detected in time, potentially leading to technical accidents such as local frequency instability and voltage exceeding limits, as well as economic losses such as regional power outages. Therefore, developing deep situational awareness technologies for microgrid clusters is crucial. Summary of the Invention

[0003] To address the aforementioned problems in related technologies, embodiments of the present invention provide a method and device for predicting a multimodal spatiotemporal power flow data storage pool network.

[0004] In a first aspect, embodiments of the present invention provide a method for predicting a multimodal spatiotemporal power flow data reservoir network, comprising: acquiring multimodal spatiotemporal power flow data, wherein the multimodal spatiotemporal power flow data includes time-series power flow data and four-dimensional coordinates of nodes in space; constructing a tree-like reservoir network model, using the tree-like reservoir network model to fuse features at different levels, and predicting the time-series power flow data to obtain future power flow data for each region; processing the four-dimensional coordinates of the nodes in space using four-dimensional fuzzy wavelet clustering to generate similar nodes; comparing the correlation between the predicted power flow data of the similar nodes, and calculating a data volatility index based on the correlation; and marking weak links in the microgrid group based on the data volatility index.

[0005] Based on the above method embodiments, the multimodal spatiotemporal current data storage pool network prediction method provided in this embodiment of the invention includes longitude, latitude, altitude and terrain in the four-dimensional coordinates of the node space.

[0006] Based on the above method embodiments, the multimodal spatiotemporal power flow data reserve pool network prediction method provided in this embodiment of the invention includes constructing a tree-structured reserve pool network model, comprising: n input nodes, m tree-structured reserve pool layers, k reserve pool nodes, and p output nodes; input u, reserve pool node state x, and output y; the input weight of the reserve pool is W_in, and the weighted adjacency matrix of the reserve pool is W_in. i,j The connection weight between the reserve pools is W. in,ij The output weight of the reserve pool is W_out.

[0007] Based on the above method embodiments, the multimodal spatiotemporal flow data reserve pool network prediction method provided in this embodiment of the invention includes a readout phase and a train phase in the training of the tree-structured reserve pool network model. In the readout phase, the reserve pool node states are updated while the weighted adjacency matrix W, input weights W_in, and output weights W_out remain unchanged. There is no backpropagation process; the states are updated based on the randomly generated weighted adjacency matrix. In the train phase, the output weights W_out are updated, and backpropagation is used to minimize the loss function. The loss function is regularized to avoid overfitting. The problem after regularization is:

[0008]

[0009] Where min represents the minimum value; L is the loss function; y target For output reference value; y o The output value is t; the duration is t; || is the norm symbol; W_out i λ represents the output weights of the i-th layer; λ is the regularization parameter.

[0010] Based on the above method embodiments, the multimodal spatiotemporal power flow data reserve pool network prediction method provided in this embodiment of the invention includes updating the reserve pool node status during the readout phase, comprising:

[0011]

[0012] Among them, R i,j Let α be the state value of the j-th reservoir in the i-th layer; i,j tanh is the dropout rate; tanh is the activation function. For the (i-1)th layer Status of each reserve pool; R n,i This represents the state of the i-th reserve pool in the n-th layer; β i,j For bias.

[0013] Based on the above method embodiments, the multimodal spatiotemporal current flow data storage pool network prediction method provided in this embodiment of the invention, wherein the four-dimensional coordinates of the node space are processed by four-dimensional fuzzy wavelet clustering to generate similar nodes, includes: fuzzifying the four-dimensional coordinates of the node space and the terrain dimension, and using a skewed large semi-gradient membership function and a skewed small semi-gradient membership function to calculate the longitude membership, latitude membership, and altitude membership of the nodes; the longitude membership calculation formula is:

[0014]

[0015] Among them, u ij For longitude membership degree; u iw latitude membership degree; u ih For altitude membership; A i B is the longitude of the i-th layer; i C represents the latitude of the i-th layer; i Let A be the elevation of the i-th layer; max The maximum longitude among all nodes; A min B is the minimum longitude among all nodes. max B is the maximum latitude among all nodes. min C is the minimum latitude among all nodes. max C represents the highest altitude among all nodes. min The lowest elevation among all nodes is determined. A hypercube region consisting of the four-dimensional coordinate system and the point (1,1,1,1) is divided into grids, with the hypercube region corresponding to the vector data in the four-dimensional coordinate system uniformly divided into grid cells of a predetermined order. Discrete wavelet transform is applied to the feature vectors generated by the data point density within the grid cells to obtain new feature vectors after low-pass filtering and smoothing. These new feature vectors are then filtered, with elements having values ​​below a threshold set to zero. Neighborhood connectivity analysis is performed on non-zero elements, clustering non-zero values ​​in adjacent regions into the same family. Grid cells of different families are numbered, and the clustering result is obtained based on the cell number where the data point is located.

[0016] Based on the above method embodiments, the multimodal spatiotemporal power flow data storage pool network prediction method provided in this embodiment of the invention, wherein marking the weak links of the microgrid group according to the data volatility index includes:

[0017]

[0018] Where softmax is the maximum flexibility function; z i Let be the sample entropy of the predicted data for node i, and be the data volatility index; e is the natural constant.

[0019] Secondly, embodiments of the present invention provide a multimodal spatiotemporal power flow data reserve pool network prediction device, comprising: a first main module for acquiring multimodal spatiotemporal power flow data, wherein the multimodal spatiotemporal power flow data includes time-series power flow data and four-dimensional coordinates of nodes; a second main module for constructing a tree-like reserve pool network model, using the tree-like reserve pool network model to fuse features at different levels, and predicting the time-series power flow data to obtain future power flow data for each region; a third main module for processing the four-dimensional coordinates of the nodes using four-dimensional fuzzy wavelet clustering to generate similar nodes; a fourth main module for comparing the correlation of predicted power flow data among the similar nodes, and calculating a data volatility index based on the correlation; and a fifth main module for marking weak links in the microgrid group based on the data volatility index.

[0020] Thirdly, embodiments of the present invention provide an electronic device, comprising:

[0021] At least one processor, at least one memory, and a communication interface; wherein,

[0022] The processor, memory, and communication interface communicate with each other;

[0023] The memory stores program instructions that can be executed by the processor. The processor calls the program instructions to execute the multimodal spatiotemporal power flow data reserve pool network prediction method provided by any one of the various implementations of the first aspect.

[0024] Fourthly, embodiments of the present invention provide a non-transitory computer-readable storage medium storing computer instructions that cause a computer to execute the multimodal spatiotemporal power flow data storage pool network prediction method provided by any of the various implementations of the first aspect.

[0025] The multimodal spatiotemporal power flow data storage pool network prediction method and device provided in this invention uses four-dimensional fuzzy wavelet clustering to process node spatial information and tree-structured storage pool network analysis to analyze time-series data, fully integrating multimodal data to improve prediction accuracy and avoid model underfitting. It enables proactive prediction of weak links in microgrids, transforming passive repair into proactive prevention, reducing the risk of power supply accidents, accurately locating weak links, reducing inspection frequency, saving operation and maintenance costs, and assisting in proactive situational awareness of microgrids. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a schematic diagram of the multimodal spatiotemporal power flow data storage pool network prediction method provided in an embodiment of the present invention;

[0028] Figure 2 A schematic diagram of the structure of the multimodal spatiotemporal power flow data storage pool network prediction device provided in an embodiment of the present invention;

[0029] Figure 3 A schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention;

[0030] Figure 4 This is a schematic diagram of wavelet clustering provided in an embodiment of the present invention;

[0031] Figure 5 This is a schematic diagram of a tree-like reservoir network model provided in an embodiment of the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form feasible technical solutions. Such combinations are not constrained by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention. If there are step numbers in the following embodiments, they are only set for ease of explanation and the order between steps is not limited. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0033] This invention provides a method for predicting multimodal spatiotemporal power flow data storage pool networks. (See also...) Figure 1The method includes: acquiring multimodal spatiotemporal power flow data, wherein the multimodal spatiotemporal power flow data includes time-series power flow data and four-dimensional coordinates of nodes; constructing a tree-like reserve pool network model, using the tree-like reserve pool network model to fuse features at different levels, predicting the time-series power flow data to obtain future power flow data for each region; processing the four-dimensional coordinates of the nodes using four-dimensional fuzzy wavelet clustering to generate similar nodes; comparing the correlation between the predicted power flow data of the similar nodes, calculating a data volatility index based on the correlation; and marking weak links in the microgrid group based on the data volatility index.

[0034] Based on the above method embodiments, as an optional embodiment, the multimodal spatiotemporal current flow data reserve pool network prediction method provided in this embodiment of the invention includes longitude, latitude, altitude and terrain in the four-dimensional coordinates of the node space.

[0035] To leverage data features across multiple time scales, a tree-structured pooling layer is used in the network to overcome the limitation of single-level features in deep pooling. By setting different pool sizes, data features at different levels can be extracted. Based on this characteristic, a tree-structured multi-pooling network is constructed to learn features at different scales of time-series data. By flexibly adjusting the output weights, multi-level features can be integrated, continuously enriching the details and thus improving prediction accuracy.

[0036] Based on the above method embodiments, as an optional embodiment, the multimodal spatiotemporal power flow data reserve pool network prediction method provided in this embodiment of the invention, wherein the construction of the tree-structured reserve pool network model includes: n input nodes, m tree-structured reserve pool layers, k reserve pool nodes, and p output nodes; input u, reserve pool node state x, and output y; the input weight of the reserve pool is W_in, and the weighted adjacency matrix of the reserve pool is W_in. i,j The connection weight between the reserve pools is W. in,ij The output weight of the reserve pool is W_out.

[0037] For details, please refer to [link / reference]. Figure 5 In the tree-structured reservoir network model, the number of input nodes is N. i The tree-structured reserve pool has n layers and N nodes. i,j The number of output nodes is N o The input is represented by u, and the state of the reserve pool node is represented by R. i,j This indicates that the output is represented by y. o This indicates that the weight input to the reserve pool is W. in,1 The weighted adjacency matrix of the reserve pool is W. i,j The connection weight between the reserve pools is W. in,ij The weight from the bottom-level reserve pool to the output is W. o,iThe number of reservoir layers can be flexibly changed based on the prediction results and application scenarios.

[0038] Based on the above method embodiments, as an optional embodiment, the multimodal spatiotemporal flow data reserve pool network prediction method provided in this embodiment of the invention includes a readout phase and a train phase in the training of the tree-like reserve pool network model. In the readout phase, the reserve pool node states are updated, while the weighted adjacency matrix W, input weight W_in, and output weight W_out remain unchanged. There is no backpropagation process, and the states are updated based on the randomly generated weighted adjacency matrix. In the train phase, the output weight W_out is updated, and backpropagation is used to minimize the loss function. The loss function is regularized to avoid overfitting. The problem after regularization is:

[0039]

[0040] Where min represents the minimum value; L is the loss function; y target For output reference value; y o The output value is t; the duration is t; || is the norm symbol; W_out i λ represents the output weights of the i-th layer; λ is the regularization parameter.

[0041] To avoid overfitting and improve the network's generalization ability, the loss function is regularized, making the final problem as follows: To avoid the influence of the initial value of the reservoir on the training and prediction results, the reservoir state needs to be initialized before training begins. A period of power flow time-series data is input to obtain the reservoir state with certain prior information. After training, the network can be used for power flow data prediction. At this time, the network input is fixed-length power flow data, resulting in corresponding equal-length reservoir state time-series data.

[0042] Based on the above method embodiments, as an optional embodiment, the multimodal spatiotemporal power flow data reserve pool network prediction method provided in this embodiment of the invention, wherein updating the reserve pool node status during the readout phase includes:

[0043]

[0044]

[0045] Among them, R i,j Let α be the state value of the j-th reservoir in the i-th layer; i,j tanh is the dropout rate; tanh is the activation function. For the (i-1)th layer Status of each reserve pool; R n,iThis represents the state of the i-th reserve pool in the n-th layer; β i,j For bias.

[0046] The training process of a reservoir network differs from that of a traditional recurrent neural network. Training is divided into two phases: readout and train. The readout phase learns the input information by updating the states of the reservoir nodes. This phase only updates the state R of the reservoir nodes. i,j Adjacency matrix W i,j Input weight W in,ij and output weight W o,i The state remains unchanged. Furthermore, there is no backpropagation process in this stage; state updates rely on a randomly generated pool-weighted adjacency matrix. Unlike traditional recurrent neural networks, pool networks use state updates rather than weight updates, and the weights remain constant throughout this stage.

[0047] It is worth noting that the randomly generated adjacency matrix W i,j There is a mathematical constraint: the spectral radius is less than 1. This step is to ensure that the network state's dependence on historical inputs decays over time, avoiding state explosion. In practice, an adjacency matrix can be randomly generated and scaled to adjust the spectral radius. Simultaneously, to reflect the sparsity of connections between reservoir nodes, the number of reservoir nodes should be as large as possible, and the adjacency matrix should be sparse. Specific node state updates are as follows (see R). i,j As shown in the diagram. Based on the state update equation, the time-series data of the reservoir state, corresponding one-to-one with the input, are obtained. The output is as shown in y. o As shown.

[0048] Based on the above method embodiments, as an optional embodiment, the multimodal spatiotemporal current data storage pool network prediction method provided in this embodiment of the invention, wherein the four-dimensional coordinates of the node space are processed by four-dimensional fuzzy wavelet clustering to generate similar nodes, includes: fuzzifying the four-dimensional coordinates of the node space and the terrain dimension, and using a skewed large semi-gradient membership function and a skewed small semi-gradient membership function to calculate the longitude membership, latitude membership, and altitude membership of the node; the longitude membership calculation formula is:

[0049]

[0050]

[0051] Among them, u ij For longitude membership degree; u iw latitude membership degree; u ih For altitude membership; A i B is the longitude of the i-th layer; i C represents the latitude of the i-th layer; iLet A be the elevation of the i-th layer; max The maximum longitude among all nodes; A min B is the minimum longitude among all nodes. max B is the maximum latitude among all nodes. min C is the minimum latitude among all nodes. max C represents the highest altitude among all nodes. min The lowest elevation among all nodes is determined. A hypercube region consisting of the four-dimensional coordinate system and the point (1,1,1,1) is divided into grids, with the hypercube region corresponding to the vector data in the four-dimensional coordinate system uniformly divided into grid cells of a predetermined order. Discrete wavelet transform is applied to the feature vectors generated by the data point density within the grid cells to obtain new feature vectors after low-pass filtering and smoothing. These new feature vectors are then filtered, with elements having values ​​below a threshold set to zero. Neighborhood connectivity analysis is performed on non-zero elements, clustering non-zero values ​​in adjacent regions into the same family. Grid cells of different families are numbered, and the clustering result is obtained based on the cell number where the data point is located.

[0052] For distributed node spatial data, four-dimensional fuzzy wavelet clustering is used for classification (e.g., Figure 4 As shown in the diagram, the four dimensions are latitude and longitude, altitude, and topography. First, the node information is fuzzified to improve the robustness of clustering. For the topography dimension, if the information source is reliable, fuzzification is not necessary; otherwise, it is required. Fuzzification uses a skewed large semi-gradient membership function and a skewed small semi-gradient membership function to calculate the longitude membership degree of the nodes, such as u. ij As shown, the latitudinal membership degree is as u iw As shown and the altitude membership degree, such as u ih As shown, after the membership degree calculation is completed, a four-dimensional membership degree vector data is obtained based on the node spatial distribution characteristics. It is worth noting that this method is not limited to the above four-dimensional data, but can also be extended to other multi-dimensional feature spaces.

[0053] The obtained four-dimensional membership vector data is uniformly segmented. Plotting this data in a four-dimensional coordinate system, the segmentation is visually represented by dividing the hypercube region enclosed by the first quadrant and the point (1,1,1,1) into 3×3×3×3 (divided into 3 blocks) or 4×4×4×4 (divided into 4 blocks) grid cells. The level of segmentation coarseness needs to balance clustering accuracy and algorithm robustness; more segmentation results in more clusters. Excessively high granularity can lead to overfitting and noise sensitivity; excessively low granularity will result in the loss of spatial similarity features. Verification shows that 3-4 order segmentation provides the optimal accuracy-efficiency balance in most scenarios.

[0054] After obtaining the grid cells, the data point density within each cell is calculated to generate the corresponding quantized feature space. The quantized feature space is four-dimensional and can be represented using a four-dimensional tensor. Feature vectors of corresponding lengths are generated based on different partitioning granularities. Then, a discrete wavelet transform is applied to the feature vectors to obtain new feature vectors after low-pass filtering and smoothing. Feature filtering is performed on the new vectors, setting elements with values ​​below a threshold to zero to eliminate noise interference. Neighborhood connectivity analysis is then performed on non-zero elements to detect whether there are clusters of non-zero values ​​in the horizontal or vertical directions; connected clusters are grouped into the same family. Different family grid cells are numbered, and the clustering result is obtained based on the cell where the data point is located. If a data point is located within a zero-value cell, it is considered that the node space data may be erroneous or that it is an isolated node.

[0055] Based on the above method embodiments, as an optional embodiment, the multimodal spatiotemporal power flow data storage pool network prediction method provided in this embodiment of the invention, wherein marking the weak links of the microgrid group according to the data volatility index includes:

[0056]

[0057] Where softmax is the maximum flexibility function; z i Let be the sample entropy of the predicted data for node i, and be the data volatility index; e is the natural constant.

[0058] After obtaining the clustering results, the volatility of the predicted data for similar nodes is compared, and nodes with greater volatility are selected and marked as weak links. The data volatility index is sample entropy, and the sample entropy z of the predicted data for node i is calculated respectively. i Then the entropy z of similar node samples i The input is fed into the softmax function to obtain the final comparison value.

[0059] The multimodal spatiotemporal power flow data reserve pool network prediction method provided in this invention uses four-dimensional fuzzy wavelet clustering to process node spatial information and tree-structured reserve pool network analysis to analyze time-series data, fully integrating multimodal data to improve prediction accuracy and avoid model underfitting. It enables proactive prediction of weak links in microgrids, transforming passive repair into proactive prevention, reducing the risk of power supply accidents, accurately locating weak links, reducing inspection frequency, saving operation and maintenance costs, and assisting in proactive situational awareness of microgrids.

[0060] The implementation of the various embodiments of this invention is based on programmed processing (i.e., software) using a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of this invention can be encapsulated into various modules. Based on this reality, and building upon the above embodiments, this invention provides a multimodal spatiotemporal power flow data storage pool network prediction device, which is used to execute the multimodal spatiotemporal power flow data storage pool network prediction method described in the above method embodiments. See also... Figure 2 The device includes: a first main module for acquiring multimodal spatiotemporal power flow data, which includes time-series power flow data and four-dimensional coordinates of nodes; a second main module for constructing a tree-like reservoir network model, using the tree-like reservoir network model to fuse features at different levels, and predicting the time-series power flow data to obtain future power flow data for each region; a third main module for processing the four-dimensional coordinates of the nodes using four-dimensional fuzzy wavelet clustering to generate similar nodes; a fourth main module for comparing the correlation of predicted power flow data among the similar nodes, and calculating a data volatility index based on the correlation; and a fifth main module for marking weak links in the microgrid cluster based on the data volatility index.

[0061] The multimodal spatiotemporal power flow data storage pool network prediction device provided in this embodiment of the invention adopts... Figure 2 Several modules within the system process node spatial information through four-dimensional fuzzy wavelet clustering, analyze time-series data using a tree-like reserve pool network, and fully integrate multimodal data to improve prediction accuracy and avoid model underfitting. This enables proactive prediction of weak links in microgrid clusters, transforming passive repairs into proactive prevention, reducing the risk of power supply accidents, and accurately locating weak links can also reduce inspection frequency, save operation and maintenance costs, and help microgrid clusters proactively perceive the situation.

[0062] It should be noted that the apparatus in the device embodiments provided by the present invention can be used not only to implement the methods in the above method embodiments, but also to implement the methods in other method embodiments provided by the present invention. The difference lies only in setting corresponding functional modules (i.e., software). Its principle is basically the same as that of the above device embodiments provided by the present invention. As long as those skilled in the art, based on the above device embodiments and referring to the specific technical solutions in other method embodiments, obtain corresponding technical means and technical solutions composed of these technical means by combining technical features, and under the premise of ensuring the practicality of the technical solution, they can improve the apparatus in the above device embodiments to obtain corresponding device-type embodiments (i.e., software) for implementing the methods in other method-type embodiments. For example:

[0063] Based on the above device embodiments, as an optional embodiment, the multimodal spatiotemporal trend data storage pool network prediction device provided in this embodiment of the invention further includes: a first sub-module, used to realize the four-dimensional coordinates of the node space including longitude, latitude, altitude and terrain.

[0064] Based on the above-described device embodiments, as an optional embodiment, the multimodal spatiotemporal power flow data reserve pool network prediction device provided in this embodiment of the invention further includes: a second submodule, used to implement the construction of the tree-like reserve pool network model, including: n input nodes, m tree-like reserve pool layers, k reserve pool nodes, and p output nodes; input u, reserve pool node state x, and output y; the input weight of the reserve pool is W_in, and the weighted adjacency matrix of the reserve pool is W_in. i,j The connection weight between the reserve pools is W. in,ij The output weight of the reserve pool is W_out.

[0065] Based on the above-described device embodiments, as an optional embodiment, the multimodal spatiotemporal flow data reserve pool network prediction device provided in this embodiment of the invention further includes: a third submodule, used to implement the training of the tree-like reserve pool network model, including a readout phase and a train phase; the readout phase updates the state of the reserve pool nodes, the weighted adjacency matrix W, the input weight W_in, and the output weight W_out remain unchanged, there is no backpropagation process, and the state is updated according to the randomly generated weighted adjacency matrix; the train phase updates the output weight W_out, uses backpropagation to minimize the loss function, and regularizes the loss function to avoid overfitting. After regularization, the problem is:

[0066]

[0067] Where min represents the minimum value; L is the loss function; y target For output reference value; y o The output value is t; the duration is t; || is the norm symbol; W_out i λ represents the output weights of the i-th layer; λ is the regularization parameter.

[0068] Based on the above-described device embodiments, as an optional embodiment, the multimodal spatiotemporal power flow data reserve pool network prediction device provided in this embodiment of the invention further includes: a fourth sub-module, used to update the reserve pool node status during the readout phase, including:

[0069]

[0070] Among them, R i,j Let α be the state value of the j-th reservoir in the i-th layer;i,j tanh is the dropout rate; tanh is the activation function. For the (i-1)th layer Status of each reserve pool; R n,i This represents the state of the i-th reserve pool in the n-th layer; β i,j For bias.

[0071] Based on the above-described device embodiments, as an optional embodiment, the multimodal spatiotemporal current data storage pool network prediction device provided in this embodiment of the invention further includes: a fifth submodule, used to implement the processing of the four-dimensional coordinates of the node space using four-dimensional fuzzy wavelet clustering to generate similar nodes, including: fuzzifying the four-dimensional coordinates of the node space and the terrain dimension, and using a skewed large semi-gradient membership function and a skewed small semi-gradient membership function to calculate the longitude membership, latitude membership, and altitude membership of the node; the longitude membership calculation formula is:

[0072]

[0073]

[0074] Among them, u ij For longitude membership degree; u iw latitude membership degree; u ih For altitude membership; A i B is the longitude of the i-th layer; i C represents the latitude of the i-th layer; i Let A be the elevation of the i-th layer; max The maximum longitude among all nodes; A min B is the minimum longitude among all nodes. max B is the maximum latitude among all nodes. min C is the minimum latitude among all nodes. max C represents the highest altitude among all nodes. min The lowest elevation among all nodes is determined. A hypercube region consisting of the four-dimensional coordinate system and the point (1,1,1,1) is divided into grids, with the hypercube region corresponding to the vector data in the four-dimensional coordinate system uniformly divided into grid cells of a predetermined order. Discrete wavelet transform is applied to the feature vectors generated by the data point density within the grid cells to obtain new feature vectors after low-pass filtering and smoothing. These new feature vectors are then filtered, with elements having values ​​below a threshold set to zero. Neighborhood connectivity analysis is performed on non-zero elements, clustering non-zero values ​​in adjacent regions into the same family. Grid cells of different families are numbered, and the clustering result is obtained based on the cell number where the data point is located.

[0075] Based on the above-described device embodiments, as an optional embodiment, the multimodal spatiotemporal power flow data storage pool network prediction device provided in this embodiment of the invention further includes: a sixth sub-module, used to implement the step of marking weak links in the microgrid group based on the data volatility index, including:

[0076]

[0077] Where softmax is the maximum flexibility function; z i Let be the sample entropy of the predicted data for node i, and be the data volatility index; e is the natural constant.

[0078] The method in this embodiment of the invention is implemented using an electronic device; therefore, it is necessary to introduce the relevant electronic device. For this purpose, this embodiment of the invention provides an electronic device, such as... Figure 3 As shown, the electronic device includes at least one processor, a communications interface, at least one memory, and a communications bus, wherein the at least one processor, the communications interface, and the at least one memory communicate with each other via the communications bus. The at least one processor can invoke logical instructions stored in the at least one memory to execute all or part of the steps of the methods provided in the foregoing method embodiments.

[0079] Furthermore, when the logical instructions in at least one of the aforementioned memories can be implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to related technologies, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various method embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0080] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0081] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. These computer software products can be stored in computer-readable storage media, such as ROM / RAM, magnetic disks, optical disks, etc., and include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0082] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Based on this understanding, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, or sometimes in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0083] It should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. Any expressions such as "predetermined threshold," "preset threshold," etc., without specifying a particular value, can be determined by those skilled in the art through simple experimentation or appropriate adjustments.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-modal spatio-temporal power flow data reservoir network prediction method, characterized in that, The method comprises the following steps: The node space four-dimensional coordinates comprise longitude, latitude, altitude and terrain. The training of the tree reserve pool network model comprises a readout stage and a train stage; the readout stage updates the reserve pool node state, and the weighted adjacency matrix W, the input weight W_in and the output weight W_out remain unchanged, there is no back propagation process, and the state is updated according to the randomly generated weighted adjacency matrix; the train stage updates the output weight W_out, adopts back propagation to minimize the loss function, and performs regularization on the loss function to avoid overfitting, and the problem after regularization is: The tree reserve pool network model is constructed, including: the number of input nodes is n, the number of tree reserve pool layers is m, the number of reserve pool nodes is k, and the number of output nodes is p; the input is u, the state of the reserve pool node is x, and the output is y; the input weight of the reserve pool is W_in, the weighted adjacency matrix of the reserve pool is , the connection weight between the reserve pools is , and the output weight of the reserve pool is W_out; The readout stage updates the reserve pool node state, which comprises: ； wherein min is the minimum value symbol; L is the loss function; is the output reference value; is the output value; t is the time length; is the norm symbol; is the output weight of the i-th layer; is the regularization parameter.

2. The multi-modal spatio-temporal power flow data reservoir network prediction method of claim 1, wherein, The four-dimensional fuzzy wavelet clustering is adopted to process the node space four-dimensional coordinates to generate similar nodes, which comprises: the node space four-dimensional coordinates and the terrain dimension are subjected to fuzzy processing, a large semi-gradient membership function and a small semi-gradient membership function are adopted to calculate the longitude membership, the latitude membership and the altitude membership of the node; the longitude membership calculation formula is: ； ； wherein, is the state value of the jth reservoir of the ith layer; is the dropout rate; tanh is the activation function; is the state of the jth reservoir of the ith-1 layer; is the state of the ith reservoir of the n layer; is the state of the ith reservoir of the n layer; is the bias.

3. The multi-modal spatio-temporal power flow data reservoir network prediction method of claim 1, wherein, According to the data volatility index, the weak link of the micro-grid group is marked, which comprises: ； ； ； wherein, is a longitude membership degree; is a latitude membership degree; is an altitude membership degree; is a longitude of the i-th layer; is a latitude of the i-th layer; is an altitude of the i-th layer; is a maximum longitude among all nodes; is a minimum longitude among all nodes; is a maximum latitude among all nodes; is a minimum latitude among all nodes; is a maximum altitude among all nodes; is a minimum altitude among all nodes; a hypercube region composed of a four-dimensional coordinate system and a (1, 1, 1, 1) point is meshed, a corresponding hypercube region of vector data in the four-dimensional coordinate system is uniformly divided into a predetermined order grid cell; a discrete wavelet transform is applied to a feature vector generated by a data point density in the grid cell, to obtain a new feature vector after low-pass filtering and smoothing; the new feature vector is screened, and elements with a value lower than a threshold value are zeroed; a neighborhood connectivity analysis is performed on non-zero elements, non-zero values in adjacent regions are clustered into the same family, different family grid cells are numbered, and a clustering result is obtained according to a cell number where a data point is located.

4. The multi-modal spatio-temporal power flow data reservoir network prediction method of claim 1, wherein, The method comprises the following steps: ； wherein softmax is a flexible maximum function; is the node The sample entropy of the prediction data is an index of data volatility; e is a natural constant.

5. A multi-modal spatio-temporal power flow data reservoir network prediction apparatus, characterized in that, The first main module is used for acquiring multi-modal space-time power flow data, the multi-modal space-time power flow data comprising time-series power flow data and node space four-dimensional coordinates; the second main module is used for constructing a tree reserve pool network model, adopting the tree reserve pool network model to fuse different levels of features, and predicting the time-series power flow data to obtain future regional power flow data; the third main module is used for adopting four-dimensional fuzzy wavelet clustering to process the node space four-dimensional coordinates to generate similar nodes; the fourth main module is used for comparing the correlation of the predicted power flow data between the similar nodes, calculating a data volatility index according to the correlation; and the fifth main module is used for marking the weak link of the micro-grid group according to the data volatility index; The node space four-dimensional coordinates comprise longitude, latitude, altitude and terrain. ​ The tree reserve pool network model is constructed, including: the number of input nodes is n, the number of tree reserve pool layers is m, the number of reserve pool nodes is k, and the number of output nodes is p; the input is u, the state of the reserve pool node is x, and the output is y; the input weight of the reserve pool is W_in, the weighted adjacency matrix of the reserve pool is , the connection weight between the reserve pools is , and the output weight of the reserve pool is W_out; The training of the tree reserve pool network model comprises a readout stage and a train stage; the readout stage updates the reserve pool node state, the weighted adjacency matrix W, the input weight W_in and the output weight W_out remain unchanged, there is no back propagation process, and the state is updated according to the randomly generated weighted adjacency matrix; the train stage updates the output weight W_out, adopts back propagation to minimize the loss function, and performs regularization on the loss function to avoid overfitting, and the problem after regularization is: ； wherein min is the minimum value symbol; L is the loss function; is the output reference value; is the output value; t is the time length; is the norm symbol; is the output weight of the i-th layer; is the regularization parameter.

6. An electronic device, comprising: Comprise: At least one processor, at least one memory and a communication interface; wherein, The processor, memory and communication interface communicate with each other; The memory stores program instructions executable by the processor, and the processor invokes the program instructions to execute the method of any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium, comprising: The non-transitory computer readable storage medium stores computer instructions, and the computer instructions enable the computer to execute the method of any one of claims 1 to 4.

Images